Judgment of Social Traits and Emotions of “Neutral” Avatars

Introduction

With the ever-increasing popularity of virtual and augmented realities, avatars have become omnipresent in research, gaming, education, and therapies, offering diverse opportunities to engage with these virtual characters. There exist several high-quality, open-source libraries that offer a wide range of avatars to the developers of virtual reality (VR) applications (e.g., Microsoft’s Rocketbox Avatar Library). The avatars in the Rocketbox Avatar Library collection were designed to “represent generic humans for the use of research and game design” and are presented as “neutral,” given that they lack facial expressions. ¹

However, social and affective evaluation involves much more than the detection of facial expressions. Ethnicity, age, gender, apparel, and other attributes or features can also impact one’s perceptions and evaluation of others.^2–5 Therefore, even if avatars are presented with a “neutral” expression, it is unclear if users view them as truly neutral. Physical attributes, such as facial features, ethnicity, and body type, may be associated with certain cultural or social stereotypes and can play a role in shaping people’s perceptions of others.^4,6 Further, apparel can signal a person’s occupation or status and can influence how the users judge the avatars. Uniforms, for example, may convey a sense of authority or professionalism. ⁷ Previous studies have found that users are able to accurately recognize the emotional expressions of human-like avatars in VR, similar to how they perceive human emotions in everyday life.^8,9 Moreover, neuroimaging data indicate the recruitment of overlapping neural circuitry for processing human-like avatars and real humans. ⁷ Therefore, it is likely that social cognitive processes that lead to stereotyping and biases in the real world transfer into the virtual world.^2,3,5

In daily life, we make fast judgments about the social traits and emotional states of others within seconds of an encounter. ¹⁰ Trait judgment, a fundamental aspect of social cognition, is the process of forming impressions based on observable behaviors, characteristics, and other available cues. Although these fast judgments can be adaptive, they can also lead to errors and stereotyping. ¹¹

The aim of this study was to examine how social traits and emotions of “neutral” avatars are perceived by users. Past research has not systematically assessed perceived social traits and emotions of presumed “neutral” avatars. We hypothesized that the avatars’ gender, ethnicity, and apparel would influence how the users evaluate the social and affective features of these virtual agents.

Methods

Measures

General Demographics Questionnaire

The REDCap survey began with a general instruction about the survey and a set of basic questions about the participant (see Table 1).

Social Trait and Emotion Judgment Task

After completing the demographic questionnaire, instructions for the Social Trait and Emotion Judgment Task were presented on the computer screen. Each trial consisted of a picture of an avatar (see Figure 1) and questions about the perceived emotion and social traits associated with the avatar. All responses were collected via mouse clicks. The order of the presentation of the avatars was randomized.

OPEN IN VIEWER

FIG. 1.

Example of a trial. This figure illustrates an example of a trial via the online REDCap questionnaire. Participants are presented with questions asking them to judge the emotion, the intensity of the emotion, and the social traits of each avatar.

First, the participant was asked to select a word that most closely matched the emotion expressed by the avatar from seven choices (happy, surprised, sad, angry, disgusted, afraid, or no emotion). This task was included to decide the most probable emotion category associated with each avatar. Then, participants were asked to rate the intensity of the attributed emotion from the previous question on a 5-point scale, ranging from “very weak” to “very strong.” These emotions were selected based on Ekman’s theory of basic emotions.^13,14

Next, participants were asked to evaluate the avatar on five social traits, which are commonly analyzed in social perception research: attractiveness, trustworthiness, friendliness, intelligence, and aggressiveness. ¹⁵ For each trait, the participant was asked to provide a rating on a 5-point scale from “not at all” to “very much.”

When all the questions were completed, participants proceeded to the next trial. Each trial featured a different avatar. Participants were able to change their responses within a trial by clicking a different selection. They were also able to return to previous trials and change their responses by clicking on the “Previous Page” button if they wished to. Participants completed all trials without a time limit.

Data analysis

We examined the effects of an avatar’s perceived gender, ethnicity, and apparel on social trait judgment ratings and emotion categories. Main effects of the avatar’s gender, ethnicity, and uniform and interactions were tested using analysis of variance. Tukey’s post hoc tests were also conducted. All analysis was conducted with JMP 17.0. The complete avatar ratings, along with images of each avatar included in the study can be found in Supplementary Data S1.

Results

Discussion

This study aimed to examine whether the avatars available through Microsoft’s Rocketbox Avatar Library, which were designed to be neutral, were truly perceived as neutral by users. We examined the responses of 225 anonymous participants to 82 avatars of different genders, ethnicities, and apparel. Overall, we found that participants did perceive these avatars to be neutral. However, our hypothesis was confirmed in that avatars’ gender, ethnicity, and apparel did influence how users evaluated their social and affective features.

Specifically, we found that females were rated higher on more positive traits, such as attractiveness, trustworthiness, and friendliness, and lower on aggression. This is consistent with previous research on sex differences in the perception of social traits in humans.^16–19

Additionally, we found that Black avatars were perceived more positively by participants, particularly in their attractiveness, friendliness, and trustworthiness. This is notable given that the majority of the participants identified as White. These findings deviate from the well-documented phenomena of in-group favoritism and cross-race effect.^20,21

Lastly, we found that avatars in uniforms were viewed as more trustworthy, intelligent, and aggressive. This is consistent with prior findings that uniforms are associated with authority, expertise, and competence.^22,23 Those in non-martial uniforms were found to be more trustworthy, intelligent, and friendly as well as less aggressive than those in martial uniforms and those without uniforms.^5,22,23

This study supports the findings that humans generally perceive the emotions of virtual avatars as similar to real-world emotions.^8,9 Our findings suggest that potential biases and stereotyping from the real world carry over into VR.^2,3,5

Our study is one of the first to use Likert-like scales to rate perceived emotion and social traits of “neutral” avatars. Our work challenges this assumption by considering that avatars may inherently possess emotional biases that influence user perceptions and interactions, regardless of this overlaying of additional emotional information. This implies that game designers and researchers need to consider these inherent biases when creating and studying avatars, as the utilization of avatars will extend beyond mere novelty and entertainment, encompassing an expansive array of applications (e.g., therapeutic interventions, digital spaces for community meetings, online shopping, educational modules).

There are limitations. First, the majority of the participants who participated in the study were White. Further, most avatars in the study were also White. A larger selection of ethnically diverse avatars in different apparel would be important for a comprehensive investigation of emotion perception and social trait judgment in the future. Additionally, it would be helpful to directly address individual differences in social cognition and exposure to avatars. Finally, in our surveys, avatars were presented as a 2D picture, rather than in 3D as they would appear in VR settings. Therefore, certain attributes such as height, weight, age, and posture were not included in the analysis. Future research should build upon the results of the current study to better understand the social and affective cues that are embedded in the visual features of these avatars. It is also important to acknowledge potential individual differences in the emotion and social cue perception capacity of the users in order to elucidate the interaction between the avatars’ visual features and participants’ social perception abilities. Given these potential confounds, it would be ideal to utilize validated and standardized emotion and social perception scales to document how these avatars are perceived by the users.

The significance of this study lies not only in unraveling the intricate interplay between design intentions and perceptual outcomes but also in highlighting the implications of “neutral” virtual representations. The impact of avatars shapes the lens through which we perceive and attribute traits and emotions to both the tangible and virtual worlds.